Comparison
oruk vs Hume AI
The honest verdict first: since June 2026 these are different products. Hume builds real-time empathic voice agents (EVI 3) and expressive TTS (Octave); it discontinued its Expression Measurement API on June 14, 2026. oruk measures speech — calibrated emotion, speaking style, and transcripts from recorded audio files. If you want an agent that talks back, use Hume. If you want to measure emotion in audio through an API, Hume no longer sells that; oruk does.
Side by side
| oruk | Hume AI | |
|---|---|---|
| Product focus (July 2026) | Measuring speech: transcription, emotion, speaking style from audio files | Building voice agents: EVI 3 speech-to-speech and Octave TTS |
| Batch emotion measurement API | Yes — POST /v1/audio/emotions, synchronous | Discontinued June 14, 2026 (Expression Measurement API) |
| Emotion output | 15 calibrated multilabel emotion scores + 16 speaking styles | EM scored 48 dimensions (prosody), plus bursts/face/text — retired |
| speech-emotion-bench accuracy | 77.6% (trained in-distribution; see methodology) | 49.6% (prosody model, measured before sunset, zero-shot) |
| Input | Prerecorded English audio files (WAV, FLAC, MP3, M4A, OGG, WebM) | Live conversational audio (EVI); text (Octave) |
| Streaming / real time | No — file-based API v1 | Yes — EVI is a real-time speech-to-speech agent |
| Pricing model | Per second of audio; emotion from $0.0060/min, analysis $0.0090/min | EVI usage-based per minute; EM was $0.0639/min before retirement |
Benchmark numbers are same-harness measurements from speech-emotion-bench (77.6% oruk Spectra vs 49.6% Hume prosody); the oruk entry is trained in-distribution — the caveat and full protocol are on the methodology page. oruk pricing is version 2026-07-11 ($0.0060/min emotion, $0.0090/min unified analysis).
Choose Hume if
- You are building a live voice agent and want speech-to-speech with expressive prosody (EVI 3).
- You need controllable, emotionally expressive text-to-speech (Octave).
- Your product needs the agent to respond in real time — oruk has no streaming API.
Choose oruk if
- You need to measure emotion or speaking style in recorded calls, meetings, or interviews.
- You want calibrated, multilabel outputs with published benchmark results rather than raw dimension scores.
- You want the transcript, emotion, and style from one request, priced per second of audio.
FAQ
- Is Hume AI still an option for measuring emotion in audio files?
- No. Hume discontinued its Expression Measurement API on June 14, 2026 and removed it from its SDKs. Hume’s current products are EVI 3 (a real-time empathic voice agent) and Octave (text-to-speech). For batch measurement of recorded audio, you need a different vendor.
- When should I choose Hume over oruk?
- Choose Hume if you are building a real-time voice agent that speaks back — EVI is a speech-to-speech product with expressive TTS, which oruk does not offer. Choose oruk if you need to measure emotion, speaking style, or transcripts from recorded audio through a REST API.
- How were the benchmark numbers measured?
- On speech-emotion-bench: 64,384 held-out clips, 7 emotion classes, one identical scoring pipeline for 64 systems; closed APIs were scored on a fixed 5,000-clip stratified subsample. The oruk entry is trained in-distribution while other systems are zero-shot — the methodology page states this caveat and provides downloadable results.
- Does oruk detect the same 48 dimensions Hume did?
- No. oruk uses a deliberately smaller space: 15 emotion labels and 16 speaking-style labels with thresholds calibrated on held-out audio, so outputs are decision-ready. A migration mapping for the common dimensions is in the migration guide.
Migration guide from Hume EM All Hume AI alternatives oruk emotion API